Hello,
If you ask me a question like:
Amine, is there any new parallel Skiplist algorithm that is more efficient than
your conccurrent algorithm that uses a distributed read-writer mutex ?
I have thought more, and i have come with a new Parallel Skiplist algorithm
that is scalable on a read mostly workload, this algorithm uses only
an mfence on the reader side, that's really cool ! and it elevates the weakness
of the distributed reader-writer mutex that becomes slower and slower on the writer side
with more an more threads, in my new concurrent Skiplist algorithm i am using one SeqlockX
between the search() and the insert() methods , you will find my SeqlockX algorithm that avoids
livelock when there is many writers here:
and i am also using one distributed reader-writer mutex between the delete() and the
search() methods...
This new algorithm is faster than my previous concurrent SkipList algorthm, but
i have made a scalability prediction for my new algorithm and i have
noticed that since the serial part of the Amdahl's law of the writer side
is a little bit expensive since the average search in a Skiplist can become a little
bit expensive, so my calculation have proved that you will not get much improvement with
this new algorithm compared to my previous algorithm, you will get around 3x improvement max,
so i have finally decided to not implement my new concurrent SkipList algorithm, so
i will stay with my current concurrent SkipList algorithm that gives a decent scalability.
Thank you,
Amine Moulay Ramdane.